CN104182625A - Electrocardiosignal denoising method based on morphology and EMD (empirical mode decomposition) wavelet threshold value - Google Patents

Abstract

The invention discloses an electrocardiosignal denoising method based on morphology and EMD (empirical mode decomposition) wavelet threshold value. The electrocardiosignal denoising method comprises the following steps: firstly, the mathematical morphology filter method is adopted to perform primary filtering on an electrocardiosignal to filter baseline drift in the electrocardiosignal, then, the EMD method is adopted to decompose the obtained electrocardiosignal and obtain a noise dominant component and a useful signal dominant component through classification, the threshold value denoising method is adopted to perform secondary filtering on the electrocardiosignal, and finally, the EMD method is adopted to reconstruct the electrocardiosignal to obtain the noise-filtered electrocardiosignal. The invention has the following remarkable effects: the method is simple and easy to implement; the morphological filter method, the EMD method and the threshold de-noising method are combined in an organic manner; compared with the traditional denoising method, the electrocardiosignal denoising method can comprehensively and effectively remove electrocardiosignal noise, keeps the useful information in the electrocardiosignal, and provides more valuable reference for change of cardiac functions and diagnosis of cardiac diseases.

Description

Denoising of ECG Signal based on morphology and EMD class wavelet threshold

Technical field

The present invention relates to biomedicine signals noise management technique field, specifically, is a kind of Denoising of ECG Signal based on morphology and EMD class wavelet threshold.

Background technology

Electrocardiosignal is one of vital sign parameter signals of people, can different aspects, accurately reflect the information of heart, is the variation of cardiac function and the diagnosis of heart disease, and the reference of a very valuable meaning is provided.

Electrocardiosignal is that a kind of randomness is strong, the non-stationary feeble signal of real-time change.Its frequency mainly concentrates on 0.05～100HZ, and energy mainly concentrates on 0.5～45HZ, and amplitude is between 0.05～5mv.The noise of electrocardiosignal mainly contains three classes: the power frequency being caused by the power supply from equipment 50HZ and higher hamonic wave is disturbed, and the baseline wander being brought by limb motion, breathing and gathering project and the myoelectricity being produced by human epidermal electromotive force, contraction of muscle disturb.The frequency of baseline wander concentrates on 0.05～10HZ, belongs to low-frequency noise; Myoelectricity interfering frequency concentrates on 5～2000HZ, belongs to high frequency noise.Noise runs through in the whole frequency domain of electrocardiosignal, brings very large difficulty to the denoising of signal.

Aspect electrocardiosignal denoising, the reasonable electrocardio noise filter having based on morphology and small echo of denoising effect, this wave filter arranges morphologic filtering device according to the feature of electrocardiosignal noise, can be extraordinary by baseline wander filtering; By wavelet transformation, the noise of HFS is effectively removed.Yet the basis function of small echo is fixed, multiresolution is constant, makes small echo lack adaptivity.And Empirical mode decomposition (Empirical Mode Decoposition, i.e. EMD) has departed from the restriction of Fourier transform, there is good adaptivity.Therefore the present invention propose a kind of based on morphology and EMD class wavelet threshold denoising method to electrocardiosignal denoising.

Summary of the invention

For the deficiencies in the prior art, the object of this invention is to provide a kind of Denoising of ECG Signal, the method has good adaptivity, can comprehensively effectively remove the noises such as baseline wander in electrocardiosignal.

For achieving the above object, the present invention explains a kind of Denoising of ECG Signal based on morphology and EMD class wavelet threshold, and its key is to carry out according to following steps:

Step 1: adopt morphologic filtering method to process the electrocardiosignal f obtaining, remove the baseline wander f in electrocardiosignal f ₁obtain signal f ₂;

Step 2: adopt empirical mode decomposition method to signal f ₂decompose, obtain K IMF component and a remaining component r _n;

Step 3: K IMF component is divided into n ₁individual noise dominant component and n ₂individual useful signal dominant component, n ₁+ n ₂=K;

Step 4: utilize threshold denoising method to n ₁individual noise dominant component carries out denoising;

Step 5: by n ₁noise dominant component after individual denoising, n ₂individual useful signal dominant component and remaining component r _ncarry out signal reconstruction, obtain the electrocardiosignal f ' after filtering noise.

As technical scheme further, the morphologic filtering method described in step 1 is carried out according to following steps:

Step 1-1: electrocardiosignal f is carried out to close-opening operation of open-closed operation He Yi road, a road simultaneously, then two-way operation result is carried out to arithmetic mean and obtain described baseline wander f ₁;

Step 1-2: by the baseline wander f of described electrocardiosignal f and step 1-1 acquisition ₁ask difference operation, obtain signal f ₂.

In conjunction with the morphological feature of baseline wander, described morphologic filtering method adopts rectilinear structure element.

As technical scheme further, the sorting technique of the component of IMF described in step 3 is as follows:

Step 3-1: the energy density E that calculates respectively K IMF component _nwith average period computing formula is respectively:

$E_n=\frac{1}{N}\underset{g=1}{\overset{N}{\mathrm{\Σ}}}{\left[{\mathrm{imf}}_n\left(g\right)\right]}^2,$

${\stackrel{\‾}{T}}_n=\frac{N}{{\mathrm{Num}}_n},$

Wherein, N is the signal length of IMF component, imf _n(g) be that n IMF divides the flow control g value of a sampled point, Num _nbe the number of maximum point in n IMF component, n=1～K;

Step 3-2: according to calculate the metewand of each IMF component, wherein, $C_n=E_n{\stackrel{\‾}{T}}_n,$ C is constant;

Step 3-3: by described metewand λ _ncompare with predetermined threshold value, work as λ _nwhile being greater than predetermined threshold value, its corresponding IMF component is useful signal dominant component, otherwise is noise dominant component.

As technical scheme further, the method for threshold denoising described in step 4 is carried out denoising according to following formula to noise dominant component:

${{\mathrm{imf}}_j}^{,}\left(i\right)=\left\{\begin{array}{cc}\mathrm{sgn}\left({\mathrm{imf}}_j\left(i\right)\right)\left(\right|{\mathrm{imf}}_j\left(i\right)\left|{-\mathrm{th}}_j\right)& \left|{\mathrm{imf}}_j\left(i\right)\right|>{\mathrm{th}}_j\\ 0& \left|{\mathrm{imf}}_j\left(i\right)\right|\≤{\mathrm{th}}_j\end{array}\right.,$

Wherein, imf _j(i) be i sampling point value of j noise dominant component, sgn () is sign function, imf _j' (i) be imf _j(i) value after denoising, th _jbe the threshold value of j noise dominant component, i=1～N, j=1～n ₁, N is the signal length of IMF component.

As technical scheme further, described threshold value th _jcomputing formula be:

${\mathrm{th}}_j={\mathrm{\σ}}_j\sqrt{2\ln N},$

Wherein, σ _jit is the standard variance of j noise dominant component.

In the present invention, first adopt mathematical morphology filter method to carry out first filtering to electrocardiosignal, baseline wander in filtered signal, then adopt EMD method to decompose electrocardiosignal, and classification draws noise dominant component and useful signal dominant component, adopt afterwards class wavelet threshold denoising method to carry out secondary filtering to electrocardiosignal, finally adopt EMD method to be reconstructed signal, obtain the electrocardiosignal after filtering noise.

Remarkable result of the present invention is: method is simple, be easy to realize, morphologic filtering method, Empirical mode decomposition and threshold denoising method are organically combined, can effectively remove the interference of the noise in electrocardiosignal, compared to traditional denoising method, have good adaptivity, the useful information in stick signal, for the variation of cardiac function and the diagnosis of heart disease provide the reference of more valuable meaning.

Accompanying drawing explanation

Fig. 1 is algorithm flow chart of the present invention;

Fig. 2 is electrocardiosignal 203 sample oscillograms;

Fig. 3 is the oscillogram of electrocardiosignal after morphologic filtering;

Fig. 4 is the oscillogram of each component after adopting EMD to decompose in the present invention;

Fig. 5 is the electrocardiosignal oscillogram after the present invention processes.

Embodiment

Below in conjunction with accompanying drawing, the specific embodiment of the present invention and principle of work are described in further detail.

Referring to accompanying drawing 1, a kind of Denoising of ECG Signal based on morphology and EMD class wavelet threshold, carries out according to following steps:

First enter step 1: the present embodiment is chosen No. 203 electrocardiogram (ECG) datas that in MIT-BIT arrhythmia cordis database, time span is 10s as pending electrocardiosignal f, its waveform as shown in Figure 2, adopt morphologic filtering method to process the electrocardiosignal f obtaining, remove the baseline wander f in electrocardiosignal f ₁obtain signal f ₂, concrete steps are as follows:

Step 1-1: electrocardiosignal f is carried out to close-opening operation of open-closed operation He Yi road, a road simultaneously, signal is done to (f ο k) k and (fk) ο k computing simultaneously, calculate afterwards the arithmetic mean of two-way operation result obtain described baseline wander f ₁;

Wherein, k is morphological structuring elements, and its length and shape directly determine the denoising performance of morphologic filtering method.Because the Main Function of filter method is cancellation baseline wander, according to the feature of electrocardiosignal and noise thereof, need to retain the characteristic wave of baseline wander, so the shape of k elects linear pattern as, its width need be greater than the width of electrocardiosignal characteristic wave.Because the width of the characteristic wave of electrocardiosignal is 54 somes left and right, No. 203 electrocardiosignals of choosing for the present embodiment, the width of k elects 72 as;

Step 1-2: by the baseline wander f of described electrocardiosignal f and step 1-1 acquisition ₁ask difference operation, i.e. f ₂=f-f ₁, the baseline wander f in cancellation original electrocardiographicdigital signal f ₁, obtain signal f ₂, signal f ₂waveform as shown in Figure 3;

Then enter step 2: adopt empirical mode decomposition method to signal f ₂decompose, obtain K IMF component and a remaining component r _n, as shown in Figure 4, in the present embodiment, decomposing gained K is 14, the imf1～imf14 shown in corresponding diagram 4 successively, and remaining component is the r in figure _n;

Enter afterwards step 3: 14 IMF components are divided into n ₁individual noise dominant component and n ₂individual useful signal dominant component, Figure 1 shows that IMFa and IMFb two classes, n ₁+ n ₂=14, sorting technique is as follows:

Step 3-1: the energy density E that calculates respectively 14 IMF components _nwith average period computing formula is respectively:

$E_n=\frac{1}{N}\underset{g=1}{\overset{N}{\mathrm{\Σ}}}{\left[{\mathrm{imf}}_n\left(g\right)\right]}^2,$

${\stackrel{\‾}{T}}_n=\frac{N}{{\mathrm{Num}}_n},$

Wherein, N is the signal length of IMF component, imf _n(g) be that n IMF divides a flow control g sampling point value, Num _nbe the number of maximum point in n IMF component, n=1～14;

Step 3-2: according to calculate the metewand of each IMF component, as shown in table 1, wherein, c is constant;

Step 3-3: because the product of white noise energy density of sound and corresponding average period is approximately constant, get constant C=2 in this example, therefore by gained metewand λ in step 3-2 _ncompare with predetermined threshold value, work as λ _nwhile being greater than predetermined threshold value, its corresponding IMF component is useful signal dominant component, otherwise is noise dominant component, thereby realizes the classification of IMF component.Be specially: in the present embodiment, predetermined threshold value is 2, as metewand λ _n≤ 2 o'clock, n IMF component was noise dominant component; Work as λ _nduring > 2, n IMF component is useful signal dominant component.

As can be seen from Table 1, metewand λ in this example _n≤ 2 noise dominant component has 11, i.e. n ₁=11, be respectively imf1～imf11; Useful signal dominant component has 3, i.e. n ₂=3, be respectively imf12～imf14;

The metewand of each IMF component of table 1

Enter afterwards step 4: utilize threshold denoising method to n ₁individual noise dominant component carries out denoising, is specially:

First, according to the feature of each noise dominant component, calculate corresponding threshold value th respectively _j, computing formula is:

${\mathrm{th}}_j={\mathrm{\σ}}_j\sqrt{2\ln N},$

Wherein, σ _jbe the standard variance of j noise dominant component, j=1～11;

Then, according to following formula, each noise dominant component is carried out to denoising:

${{\mathrm{imf}}_j}^{,}\left(i\right)=\left\{\begin{array}{cc}\mathrm{sgn}\left({\mathrm{imf}}_j\left(i\right)\right)\left(\right|{\mathrm{imf}}_j\left(i\right)\left|{-\mathrm{th}}_j\right)& \left|{\mathrm{imf}}_j\left(i\right)\right|>{\mathrm{th}}_j\\ 0& \left|{\mathrm{imf}}_j\left(i\right)\right|\≤{\mathrm{th}}_j\end{array}\right.,$

Wherein, imf _j(i) be i sampling point value of j noise dominant component, sgn () is sign function, imf _j' (i) be imf _j(i) value after denoising, th _jbe the threshold value of j noise dominant component, i=1～N, j=1～n ₁, N is the signal length of IMF component.

Finally enter step 5: by the noise dominant component after 11 denoisings, 3 useful signal dominant component and remaining component r _ncarry out signal reconstruction, obtain the electrocardiosignal f ' after filtering noise, its waveform as shown in Figure 5.

In the present invention, first adopt morphologic filtering method to carry out first filtering to electrocardiosignal f, the baseline wander f in filtered signal ₁, obtain signal f ₂, then adopt EMD method to signal f ₂decompose, according to metewand, classification draws noise dominant component and useful signal dominant component, adopt afterwards threshold denoising method to carry out secondary filtering to noise dominant component, finally adopt EMD method to carry out signal reconstruction, obtain the electrocardiosignal f ' after filtering noise.By the present invention, can effectively remove the interference of noise in electrocardiosignal comprehensively, compared to traditional denoising method, there is good adaptivity, the useful information in stick signal.

Claims (6)

1. the Denoising of ECG Signal based on morphology and EMD class wavelet threshold, is characterized in that carrying out according to following steps:

Step 1: adopt morphologic filtering method to process the electrocardiosignal f obtaining, remove the baseline wander f in electrocardiosignal f ₁obtain signal f ₂;

Step 2: adopt empirical mode decomposition method to signal f ₂decompose, obtain K IMF component and a remaining component r _n;

Step 3: K IMF component is divided into n ₁individual noise dominant component and n ₂individual useful signal dominant component, n ₁+ n ₂=K;

Step 4: utilize threshold denoising method to n ₁individual noise dominant component carries out denoising;

Step 5: by n ₁noise dominant component after individual denoising, n ₂individual useful signal dominant component and remaining component r _ncarry out signal reconstruction, obtain the electrocardiosignal f ' after filtering noise.

2. the Denoising of ECG Signal based on morphology and EMD class wavelet threshold according to claim 1, is characterized in that: the morphologic filtering method described in step 1 is carried out according to following steps:

Step 1-1: electrocardiosignal f is carried out to close-opening operation of open-closed operation He Yi road, a road simultaneously, then two-way operation result is carried out to arithmetic mean and obtain described baseline wander f ₁;

Step 1-2: by the baseline wander f of described electrocardiosignal f and step 1-1 acquisition ₁ask difference operation, obtain signal f ₂.

3. the Denoising of ECG Signal based on morphology and EMD class wavelet threshold according to claim 1 and 2, is characterized in that: described morphologic filtering method adopts rectilinear structure element.

4. the Denoising of ECG Signal based on morphology and EMD class wavelet threshold according to claim 1, is characterized in that: the sorting technique of the component of IMF described in step 3 is as follows:

Step 3-1: the energy density E that calculates respectively K IMF component _nwith average period computing formula is respectively:

$E_n=\frac{1}{N}\underset{g=1}{\overset{N}{\mathrm{\Σ}}}{\left[{\mathrm{imf}}_n\left(g\right)\right]}^2,$

${\stackrel{\‾}{T}}_n=\frac{N}{{\mathrm{Num}}_n},$

Wherein, N is the signal length of IMF component, imf _n(g) be that n IMF divides the flow control g value of a sampled point, Num _nbe the number of maximum point in n IMF component, n=1～K;

Step 3-2: according to calculate the metewand of each IMF component, wherein, $C_n=E_n{\stackrel{\‾}{T}}_n,$ C is constant;

Step 3-3: by described metewand λ _ncompare with predetermined threshold value, work as λ _nwhile being greater than predetermined threshold value, its corresponding IMF component is useful signal dominant component, otherwise is noise dominant component.

5. the Denoising of ECG Signal based on morphology and EMD class wavelet threshold according to claim 1, is characterized in that: the method for threshold denoising described in step 4 is carried out denoising according to following formula to noise dominant component:

${{\mathrm{imf}}_j}^{,}\left(i\right)=\left\{\begin{array}{cc}\mathrm{sgn}\left({\mathrm{imf}}_j\left(i\right)\right)\left(\right|{\mathrm{imf}}_j\left(i\right)\left|{-\mathrm{th}}_j\right)& \left|{\mathrm{imf}}_j\left(i\right)\right|>{\mathrm{th}}_j\\ 0& \left|{\mathrm{imf}}_j\left(i\right)\right|\≤{\mathrm{th}}_j\end{array}\right.,$

Wherein, imf _j(i) be i sampling point value of j noise dominant component, sgn () is sign function, imf _j' (i) be imf _j(i) value after denoising, th _jbe the threshold value of j noise dominant component, i=1～N, j=1～n ₁, N is the signal length of IMF component.

6. the Denoising of ECG Signal based on morphology and EMD class wavelet threshold according to claim 5, is characterized in that: described threshold value th _jcomputing formula be:

${\mathrm{th}}_j={\mathrm{\σ}}_j\sqrt{2\ln N},$

Wherein, σ _jit is the standard variance of j noise dominant component.